A guest post by Bogdan Pasaniuc [@bpasaniuc] on his paper with coauthors: Spatial localization of recent ancestors for admixed individuals by Wen-Yun Yang, Alexander Platt, Charleston Wen-Kai Chiang, Eleazar Eskin, John Novembre, Bogdan Pasaniuc. bioRxived here.

Geographic localization based on genetic data has received much attention recently. Here we present a preprint that aims to address one of the drawbacks of existing approaches. As opposed to existing works that typically make a very strong assumption that all recent ancestors come from the same location on a map, we seek to infer multiple locations for a given individual corresponding to its ancestors. That is, our approach uses genetic data from a given individual to localize on the map its recent ancestors several generations ago (e.g. grandparents).

To accomplish this we approximate the admixture process (i.e. mixing of genetic variants from different sources) in a genetic-geographic continuum. We view the mixed ancestry genome as being generated from several locations on a map (corresponding to its recent ancestors) and model the mosaic structure of local ancestries across the genome through an admixture HMM. We link geography to the admixture process by allowing allele frequencies at every site in the genome to vary across geography according to a logistic gradient function (as in SPA[1]); the complete model is an admixture HMM for a genotype-specific pair of ancestral locations on the map.

As the number of generations since admixture increases the total number of ancestors to localize increases dramatically making the inference infeasible (http://gcbias.org/2013/11/11/how-does-your-number-of-genetic-ancestors-grow-back-over-time/). To account for this, we limit the number of different “ancestry locations” that contribute to admixture to a small constant, each with varying amount of contribution. We devise efficient algorithms to make inferences in our model and show that accuracy decreases with number of locations to infer, with number of generations in the admixture and with geographic distance among ancestors. For example, SPAMIX can localize the grandparents of the POPRES[2] individuals with multiple sub-continental European ancestries within 470Km of their reported locations.

As with all methods, limitations do exist and we outline several here. We use logistic gradient functions to relate geography to genetics and investigating more complex functions may prove fruitful. We developed an efficient algorithm for producing point estimates for location and locus-specific ancestry; in some cases a probabilistic output may be desired. Finally, our approach models admixture-LD and assumes no background LD; more involved procedures to model background LD (such as the one we proposed [3]) is an interesting area of research.

1. Yang, Wen-Yun, et al. “A model-based approach for analysis of spatial structure in genetic data.” Nature genetics 44.6 (2012): 725-731.
2. Nelson, Matthew R., et al. “The population reference sample, POPRES: a resource for population, disease, and pharmacological genetics research.” The American Journal of Human Genetics 83.3 (2008): 347-358.
3. Baran, Yael, et al. “Enhanced localization of genetic samples through linkage-disequilibrium correction.” The American Journal of Human Genetics 92.6 (2013): 882-894.